Adjacency list implementation in C++ - c++

Question

I am looking for a concise and precise adjacency list representation of a graph in C++. My nodes are just node ids. Here is how I did it. Just want to know what experts think about it. Is there a better way?
This is the class implementation (nothing fancy, right now don't care about public/private methods)
#include <iostream>
#include <vector>
#include <fstream>
#include <sstream>
using namespace std;
class adjList {
public:
int head;
vector<int> listOfNodes;
void print();
};
void adjList :: print() {
for (int i=0; i<listOfNodes.size(); ++i) {
cout << head << "-->" << listOfNodes.at(i) << endl;
}
}
class graph {
public:
vector<adjList> list;
void print();
};
void graph :: print() {
for (int i=0; i<list.size(); ++i) {
list.at(i).print();
cout << endl;
}
}
My main function parses an input file line by line. Where each line is interpreted as following:
<source_node> <node1_connected_to_source_node> <node2_connected_to_source_node <node3_connected_to_source_node> <...>
Here is the main:
int main()
{
fstream file("graph.txt", ios::in);
string line;
graph g;
while (getline(file, line)) {
int source;
stringstream str(line);
str >> source;
int node2;
adjList l;
l.head = source;
while (str >> node2) {
l.listOfNodes.push_back(node2);
}
g.list.push_back(l);
}
file.close();
g.print();
getchar();
return 0;
}
I know I should add addEdge() function inside adjList class instead of directly modifying its variable from main() however, right now I just wonder about the best structure.
EDIT:
There is one shortcoming in my approach. For a complicated graph with large number of nodes, node will indeed be a struct/class and in that case I will be duplicating values by storing the whole object. In that case I think I should use pointers. For example for an undirected graph, I will be storing copies of node objects in the adjList (connection between node 1 and 2 means 1's adjacency list will have 2 and vice versa). I can avoid that by storing pointers of node objects in the adjList instead of the whole object. Check the dfs implementation which get benefited by this approach. There I need to insure that each node gets visited only once. Having multiple copies of the same node will make my life harder. no?
In this case my class definitions will change like this:
#include <iostream>
#include <vector>
#include <fstream>
#include <sstream>
#include <map>
using namespace std;
class node {
public:
node() {}
node(int id, bool _dirty): node_id(id), dirty(_dirty) {}
int node_id;
bool dirty;
};
class adjList {
public:
node *head;
vector<node*> listOfNodes;
void print();
~adjList() { delete head;}
};
void adjList :: print() {
for (int i=0; i<listOfNodes.size(); ++i) {
cout << head->node_id << "-->" << listOfNodes.at(i)->node_id << endl;
}
}
class graph {
public:
vector<adjList> list;
void print();
void dfs(node *startNode);
};
void graph::dfs(node *startNode) {
startNode->dirty = true;
for(int i=0; i<list.size(); ++i) {
node *stNode = list.at(i).head;
if (stNode->node_id != startNode->node_id) { continue;}
for (int j=0; j<list.at(i).listOfNodes.size(); ++j) {
if (!list.at(i).listOfNodes.at(j)->dirty) {
dfs(list.at(i).listOfNodes.at(j));
}
}
}
cout << "Node: "<<startNode->node_id << endl;
}
void graph :: print() {
for (int i=0; i<list.size(); ++i) {
list.at(i).print();
cout << endl;
}
}
And this is how I implemented main() function. I am using a map<> to avoid duplication of objects. Creating a new object only when its not defined earlier. Checking existence of an object by its id.
int main()
{
fstream file("graph.txt", ios::in);
string line;
graph g;
node *startNode;
map<int, node*> nodeMap;
while (getline(file, line)) {
int source;
stringstream str(line);
str >> source;
int node2;
node *sourceNode;
// Create new node only if a node does not already exist
if (nodeMap.find(source) == nodeMap.end()) {
sourceNode = new node(source, false);
nodeMap[source] = sourceNode;
} else {
sourceNode = nodeMap[source];
}
adjList l;
l.head = sourceNode;
nodeMap[source] = sourceNode;
while (str >> node2) {
// Create new node only if a node does not already exist
node *secNode;
if (nodeMap.find(node2) == nodeMap.end()) {
secNode = new node(node2, false);
nodeMap[node2] = secNode;
} else {
secNode = nodeMap[node2];
}
l.listOfNodes.push_back(secNode);
}
g.list.push_back(l);
startNode = sourceNode;
}
file.close();
g.print();
g.dfs(startNode);
getchar();
return 0;
}
SECOND EDIT
After Ulrich Eckhardt suggestion to put adjacency list in node class, here is what I think is a better data structure to store a graph and perform dfs(), dijkstra() kind of operations. Please note that adjacency list is merged in node class.
#include <iostream>
#include <vector>
#include <fstream>
#include <sstream>
#include <map>
using namespace std;
class node {
public:
node() {
}
node(int id, bool _dirty): node_id(id), dirty(_dirty) {
//cout << "In overloaded const\n";
}
int node_id;
bool dirty;
vector<node*> listOfNodes;
};
class graph {
public:
vector<node*> myGraph;
void dfs(node* startNode);
};
void graph::dfs(node* startNode) {
startNode->dirty = true;
for (int j=0; j<startNode->listOfNodes.size(); ++j) {
if (!startNode->listOfNodes.at(j)->dirty) {
dfs(startNode->listOfNodes.at(j));
}
}
cout << "Node: "<<startNode->node_id << endl;
}
Can we do better than this?

Answer 1

There are a few things that could be improved, but in general your approach is reasonable. Notes:
You are using int as index into a container, which will give you warning from some compilers, because the size of a container could exceed the size representable as int. Instead, use size_t.
Rewrite your for (int i=0; i<list.size(); ++i) to for(size_t i=0, size=list.size(); i!=size; ++i). Using != instead of < will work with iterators. Reading and storing the size once makes it easier to debug and possibly even more efficient.
Inside the loop to print, you have list.at(i).print();. The list.at(i) will verify the index is valid and raise an exception when not. In this very simple case, I am sure that the index is valid, so using list[i] instead is faster. Also, it implicitly documents that the index is valid and not that you expect it to be invalid.
The print() functions should be constant.
I don't understand what the int head is. Is this some kind of ID for the node? And isn't the ID simply the index inside graph::list? If it is the index, you could compute that on demand using the address of the element minus the address of the first element, so there's no need to store it redundantly. Also, consider validating that index when reading, so you don't have any edges going to a vertex that doesn't exist.
If you don't care about encapsulation on a node-level (which is reasonable!), you could also make this a struct, which saves some typing.
Storing pointers instead of indices is tricky but could improve speed. The problem is that for reading, you might need a pointer to a vertex that doesn't exist yet. There is a hack that allows doing that without using additional storage, it requires first storing the indices in the pointer values (using reinterpret_cast) and after reading, making a second pass on the data where you adjust these values to the actual addresses. Of course, you can also use the second pass to validate that you don't have any edges going to vertices that don't exist at all (which is a place where the at(i) function becomes useful) so this second pass to verify some guarantees is a good thing anyway.
On explicit request, here's an example for how to store an index in a pointer:
// read file
for(...) {
size_t id = read_id_from_file();
node* node_ptr = reinterpret_cast<node*>(id);
adjacency_list.push_back(node_ptr);
}
/* Note that at this point, you do have node* that don't contain
valid addresses but just the IDs of the nodes they should finally
point to, so you must not use these pointers! */
// make another pass over all nodes after reading the file
for(size_t i=0, size=adjacency_list.size(); i!=size; ++i) {
// read ID from adjacency list
node* node_ptr = adjacency_list[i];
size_t id = reinterpret_cast<size_t>(node_ptr);
// convert ID to actual address
node_ptr = lookup_node_by_id(id);
if(!node_ptr)
throw std::runtime_error("unknown node ID in adjacency list");
// store actual node address in adjacency list
adjacency_list[i] = node_ptr;
}
I'm pretty sure that this works in general, though I'm not 100% sure if this is guaranteed to work, which was why I'm reluctant to post this here. However, I hope this also makes clear why I'm asking what exactly "head" is. If it is really just the index in a container, there is little need for it, neither inside the file nor in memory. If it is some kind of name or identifier for a node that you retrieved from a file, then you absolutely need it, but then you can't use it as index, the values there could as well start their IDs with 1 or 1000, which you should catch and handle without crashing!